Who Owns the Seam Between Your Web App and Your Mobile App?
Who Owns the Seam Between Your Web App and Your Mobile App?
Two vendors, one product. The web company owns the website. The mobile company owns the app. Both contracts are clear. Both teams are competent.
So who owns the space between them?
Nobody. And that is not an oversight — it is structural. Each vendor is measured on their own deliverable. The seam belongs to neither, so neither maintains it, and it fills with everything that was easier to duplicate than to negotiate.
What accumulates in an unowned space
Duplicated rules. Pricing, entitlements, validation — written twice, tested separately, drifting apart. The customer finds out first, usually by noticing the app and the website disagree about a number.
API design by whoever argued hardest. Which in practice means it is convenient for the browser and hostile to a phone on a train.
Double discovery. Two requirements phases, the same stakeholder interviews, billed twice.
Defect ping-pong. Every cross-surface bug consumes two triage cycles before anyone writes code.
Two analytics schemes. No honest cross-platform funnel, so product decisions get made on instinct and described as data-driven.
On the split-vendor projects our team has taken over, that seam runs fifteen to twenty-five percent of total engineering spend. It never appears as a cost. It appears as velocity slightly below plan, which nobody can cut from a budget.
One decision removes most of it
Put the meaning of the system in one place.
A single typed package: entities, state machines, validation, pricing, permissions. It imports nothing from any interface framework. It has tests that run in milliseconds without a browser or a simulator. It has one named owner and a version number.
Both clients consume it. Neither client contains a business rule.
Above it, diverge completely. Native navigation has no meaningful relationship to browser history; gesture handling has none to hover states. Share the meaning, not the pixels. The failed cross-platform projects of the last decade did exactly the reverse — shared presentation, duplicated meaning — which is why they felt subtly wrong on every platform.
Why this matters more now than it did
Assistants are genuinely strong at the derivative work of a second platform: generating typed clients from a schema, porting a validated form between surfaces, producing a test matrix for a state machine that already exists. That is real leverage on the work that used to be tedious.
The failure mode is quiet. Ask an agent to add a feature and it implements the rule where it happens to be working, because it does not know your domain package exists. The code compiles. The tests pass. Review approves it, because the diff looks fine and there were eleven others that day.
Six months of that and the rule is in four places again — the exact problem the shared core existed to prevent, reintroduced faster than any human could have managed.
So the pattern stopped being a nice-to-have when code generation got good. It became the thing holding the system together.
The diagnostic
Look at where your defects originate.
Mostly presentation issues — layout, navigation, edge-case rendering — means the core is genuinely shared and doing its job.
Business-logic bugs appearing on one surface only means the core is not shared. It is copied. Copies diverge on their own schedule, and you find out from a support ticket.
If you are hiring right now
One question does most of the work: where will our business rules live, and how would you prove it to me six months from now?
Precision means they have built this. Vagueness means you will own the seam yourself, whether or not anyone tells you so.
Full version, including the nine questions that expose a weak vendor and what the first ninety days should produce: web and mobile app development company
Frequently Asked Questions
We already have two vendors — is it too late?
No. Take ownership of the domain package yourself and have both vendors consume it. That one move removes most of the duplication without renegotiating either contract.
Does this need a specific technology stack?
No. It needs an import boundary and the discipline to hold it. Everything else is an implementation detail.
What if we only have one platform today?
The package still pays for itself in testability, and it makes a second surface cheap if you ever want one.
How do we stop rules leaking into the clients?
Mechanically. A lint rule banning interface framework imports in the package, and a CI check that flags business constants appearing in client code. Reviewer attention does not scale; lint rules do.
Who should own the package?
One named person with veto power over where logic lives. Shared ownership of this specific thing reliably means none.
What is the fastest way to test a vendor?
A paid six-week vertical slice: one real workflow, both surfaces, real infrastructure. It tells you what no reference call will.
Ese dato de que el seam se come entre 15 y 25% del gasto de ingeniería me voló la cabeza, aunque tiene todo el sentido: es costo que aparece como "velocidad un poco por debajo del plan" y nunca lo podés cortar del presupuesto. Lo del "share the meaning, not the pixels" es exactamente lo que hicieron al revés los proyectos cross-platform que fallaron. ¿Ese paquete tipado único lo pensás como monorepo compartido o lo publicarías como librería versionada aparte?